Neutron Scattering Analysis of Bacterial Lipopolysaccharide Phase Structure CHANGES AT HIGH pH*

The aggregate structure of lipopolysaccharide iso- lated from an Re strain of Escherichia coli was exam-ined at different pH values using small angle neutron scattering. At pH values of 6 and 7.4, angle-averaged scattering of the sodium salt of this isolate was con- sistent with randomly coiled tubular micelles approximately 100 A in diameter. At pH 9.1, however, Kratky analysis of the scattering data was distinctly different and consistent with pairing of uniform tubular micelle sections of length 1440 and 110 A in diameter. Contrast variation measurements of the micelles yielded an average micellar weight of the sample at pH 9.1 of approximately 1.1 1 X lo' daltons and suggested that the aggregates were tubular micelles of size and length similar to that derived from the scattering intensity data. Anisotropic scattering patterns of samples under shear indicated a rigidification of the micelles as the pH was increased to 9.1 and the temperature decreased from 25 to 10 "C. The rotational diffusion constants deduced from the observed shear anisotropy indicate that the structure at pH 9.1 must have smallest and largest dimensions which differ by at least an order of magnitude, ruling out spherical or moderately ellipsoidal structures. Analysis of the shear rate needed to induce anisotropic scattering indicated that the stiffness leagth of the micelles at pH 9.1 was approximately 1000 A and decreased at higher and lower pH values.

The aggregate structure of lipopolysaccharide isolated from an Re strain of Escherichia coli was examined at different pH values using small angle neutron scattering. At pH values of 6 and 7.4, angle-averaged scattering of the sodium salt of this isolate was consistent with randomly coiled tubular micelles approximately 100 A in diameter. At pH 9.1, however, Kratky analysis of the scattering data was distinctly different and consistent with pairing of uniform tubular micelle sections of length 1440 and 110 A in diameter. Contrast variation measurements of the micelles yielded an average micellar weight of the sample at pH 9.1 of approximately 1.1 1 X lo' daltons and suggested that the aggregates were tubular micelles of size and length similar to that derived from the scattering intensity data. Anisotropic scattering patterns of samples under shear indicated a rigidification of the micelles as the pH was increased to 9.1 and the temperature decreased from 25 to 10 "C. The rotational diffusion constants deduced from the observed shear anisotropy indicate that the structure at pH 9.1 must have smallest and largest dimensions which differ by at least an order of magnitude, ruling out spherical or moderately ellipsoidal structures. Analysis of the shear rate needed to induce anisotropic scattering indicated that the stiffness leagth of the micelles at pH 9.1 was approximately 1000 A and decreased at higher and lower pH values.
A variety of physical techniques has been used to measure the architecture and conformation of isolated LPS' from smooth and rough strains of enteric bacteria (1)(2)(3)(4)(5)(6). While some investigators suggest that under most conditions extracted LPS exists as bilayers (1,5), others have indicated that, under certain conditions, LPS may form nonlamellar aggregates (2-4,6). Recent studies have suggested that ReLPS may form inverted micellar structures (3,7), and that LPS from smooth strains (8)  the pH-dependent changes in the architecture of LPS aggregates, the sodium salt of LPS from an Re mutant strain was analyzed using small angle neutron scattering. The results indicate that at neutral pH the aggregates exist as randomly coiled tubular micelles, but at pH 9.1 the micelles are paired. Analysis of the samples under shear indicates that the micelles stiffen as the pH is increased from 7.4 to 9.1 and the temperature is lowered below 25 "C. We propose that upon elevating the pH to 9.1 the randomly coiled micelles stiffen and loop back on themselves to form closed tubular micelles.

Lipopolysaccharide Preparation
Escherichia coli strain D21f2 (an ReLPS producing strain, 9) was grown at 37 "C in nutrient broth, and the LPS was extracted using phenol/chloroform/petroleum ether as previously described (10). The sample was further purified by sedimenting the aggregate several times at 78,000 X g for 120 min and resuspending the sample in distilled water. The sodium salt of LPS (NaLPS) was obtained by dialyzing the extract against five changes of 10 mM sodium ethylenediaminetetraacetic acid, pH 7.0, at 4'C, followed by extensive dialysis against distilled water. Elemental analysis of the sodium salt was carried out as described previously (10). The levels of the LPS were quantitated using the thiobarbituric acid assay for 3-keto-~manno-octalosonic acid (11) and by the dry weight of lyophilized samples. The samples were shown to contain less than 1% protein (by weight) using the Lowry method (12). The levels of nucleic acid were determined to be less than 1% by measuring the absorption at 260 nm. For neutron-scattering studies, lyophilized samples were weighed and suspended at 10 mg/ml in either 50 mM sodium phosphate (pH 6.0 and 7.4) or sodium borate (pH 9.1 and 9.5) containing the appropriate level of D20. The LPS isolate was identical for all of the pH studies. D20 was redistilled prior to its use in the buffers. All buffers were adjusted to the indicated pH using a combination glass electrode and pH values were not corrected for the presence of D20.

Neutron-scattering Studies
Static-scattering measurements were done using 1-mm path length cylindrical quartz cells and measured at 25'C. For scattering of samples under shear, a concentric cylindrical, water-jacketed quartz cell with a total optical path of 1 mm was used (13). The exterior cylinder of the cell was motor driven, and the rate of rotation was monitored. At the maximum speed of 3000 rpm, a shear rate G = 18,500 s-l was obtained.
Neutron-scattering measurements were performed at an incident wavelength X = 4.75 A on the 30-m small angle neutron-scattering instrument of the National Center for Small Angle Scattering Research at Oak Ridge National Laboratory. Slits diameters of 12 mm at the sample and 25 mm at the source were used in conjunction with a sample-to-detector distance of 12 m. Data were corrected for sample transmission, cell and spectrometer background, solvent scattering, well-established standard techniques (14). (The resulting statistical detector efficiency, and incoherent scattering from the sample, using errors are marked as vertical bars on all intensity data presented.) Absolute intensity was calibrated using an irradiated aluminum scat-5100 tering standard (15). A constant systematic error in the absolute scale, estimated to be +5%, may be present in all runs; statistical errors in the relative intensities are generally f 2 % . In the present study, the contrast of the sample was varied by the H,O/D,O ratios in the solvent. For other analyses of sample shape by static, angle-average scattering as well as asymmetric scattering under shear, the sample was suspended in 81% DzO since the transmission is higher in DzO and the signal to noise is substantially higher than in pure HzO.
Guinier Analysis-The angle-averaged scattering data can be used to determine the radius of gyration of a particle. In the low k region, Guinier's law states: Z(k) = I(0)exp(-kZR,2/3) where Z(0) is the forward scatter, k = (4~/X)sinO, and RB is the radius of gyration. Z(0) and RB were determined by a plot of In Z(k) uersus kZ. Z(0) is related to the aggregate's particle weight and its partial specific volume.
Kratky Analysis-Data collected to large k values were interpreted by the method of Kratky and Pitz (16). This is a search method which assumes that the shape of the particle can be approximated by simple geometric objects. The scattering curve of kzZ(k) as a function of k is then compared to different models and analyzed for the best fit. To test the static scattering data, particles of randomly oriented coils, uniform cylinders, and paired uniform cylinders were considered; the expected Kratky plot of scatter from such particles was calculated.' The shear data was resolved into scatter parallel, I ( k It), and perpendicular, I ( k l ) , to the fluid flow, and tested against models of oriented cylinders (13). The calculations for the expected scatter from such models under shear are also given.' Rotational Diffusional Analysis of Samples under Shear-Broersma (17) has shown that for a cylinder of radius, a, and stiffness semilength, ! , the diffusion coefficient, D,, would be related to length and diameter by the following relationship: where q is the solvent viscosity, s = log(2f/a) and t = 1.57 -7(0.28 l/s)'. We would expect the onset of anisotropy when G / D , 1.
Thus, the shear rate corresponding to the onset of anisotropic scattering was used to estimate the stiffness length of the micelles. This is the average length of the aggregate without a bend. For these analyses, the diameters of the micelles were assumed to be between 100 and 160 A, consistent with the values obtained in the staticscattering experiments and with the known monomer dimensions.
The estimated values were then refined by comparison with detailed model calculations' ("Results").
An important feature of this type of experiment is that the very existence of anisotropy in the scatting, under the mild shear conditions used, rules out entire classes of possible structural models. In the present case, any proposed model must lead to a rotational diffusion coefficient corresponding to the observed value -2000 s", regardless of any other features. Structures with spherical symmetry are thus ruled out a priori, and lamellar structures would not show anisotropic scattering in the particular shear geometry used. Either cylindrical or ellipsoidal shapes are possible, provided their axial ratios are of order 1O:l; at this axial ratio, the two shapes are practically indistinguishable, and we have used cylinders for the model calculations.

RESULTS
Chemical Characterization of Isolated LPS-The chemical structure of the LPS from E. coli strain D21f 2 has been analyzed by several investigators (18,19) and is shown in Fig.  1. The level of monophosphate and pyrophosphate at the reducing end of the lipid A head group is variable and presumably depends on the growth conditions (20). Thus, analysis of the phosphate content of the NaLPS was carried out using plasma emission spectroscopy (10). Elemental analysis showed that the sample contained approximately 2 phosphorus atoms/LPS molecule (Table I). Thus, the reducing end is  substituted mainly with monophosphate. Elemental analysis also indicated that the isolate was neutralized with Na+ and had only very low levels of contaminating multivalent metal cations. The thiobarbituric acid assay indicated that the sample contained greater than one 3-keto-D-manno-octalosonic acid unit. We assume that the sample contained two 3-keto-D-manno-octa~osonic acid/LPS, but under the mild acid hydrolysis conditions used, incomplete release of the sugar resulted in incomplete detection of the residue. Since the sample was shown to be free of protein and nucleic acid contamination, we used the dry weight of lyophilized LPS to quantitate the amount of material.
Angle-aueruged Static Scattering-The angle-averaged scattering patterns of the NaLPS at pH 6.0, 7.4, and 9.5 did not differ substantially (Fig. 2). Data at these three pH values yield linear log I ( k ) uersus log k plots, with slopes close to the value (-2) characteristic of a random coil structure. None of the data shown in Fig. 2 yields a linear Guinier plot (log I ( k ) uersus k 2 ) , but estimates of the largest and smallest dimensions can be made from the asymptotic low and high k Guinier slopes. Except at pH 9.1, this yields charaFteristic small and large radii of gyration of order 47 and 280 A, and the data are consistent with "floppy" tubular micelles in a loosely coiled configuration. At pH 9.1, however, the data differs qualitatively from that at lower or higher pH, with the existence of a k-range in which the slope is -1, indicative of very elongated particles.
The intensity of scatter at pH 6.0, 7.4, 9.1, and 9.5 was fitted to modeled data. The I(k) uersus k and Kratky plots for different shaped particles were calculated.* A random coiled particle would be expected to give a straight line Kratky plot; i.e. k21(k) would be approximately constant, and this was found to be the case for the scattering data for the NaLPS at pH 6.0, 7.4, and 9.5. In contrast, at pH 9.1, the Kratky plot was unique and showed pronounced structure (Fig. 3). Anal- ysis of several models indicates that the scatter was best fit by particles which were raqdomly oriented parallel pairs of unitorm tubes 1440 f 130 A long and spaced apart by 40 f 10 A, surface-to-surface (Fig. 3, dotted line), when the tube diameter was taken as 110 A (see below).
Size and volume measurements of the particles of NaLPS at pH 9.1 were also calculated from the contrast variation measurements.2 The I ( 0 ) data at five H20:D20 ratios allowed for the calculation of volume2. From the structure of the molecule (Fig. l), the neutron-scattering amplitude density and formula weights were calculated. (Note that the dry formula weight is obtained in such a measurement.) Using the D20:H20 ratio that gave Z(0) = 0 and the slope of the contrast variation plot (Fig. 4), the aggregation number of the particle was calculated to be 4900 k 300. Assuming the particles were tubular micelles with a diameter of between 100 and 110 A, the lengtho of the micelles waso calculated to be between 1440 k 130 A and 1740 k 150 A, and the mean density of the dry particle was found to be 1.33 f 0.08 g/cm3. This value is very similar to the densities, 1.25 g/cm3 (21) and 1.37 g/cm3 (22), calculated from the partial specific volumes reported for ReLPS from Salmonella typhimurium and Salmonella minnesota, respectively.
Anisotropic Scattering-Anisotropic scattering patterns of NaLPS at pH 7.5, 9.1, and 9.5 were taken as a function of applied shear, G (0 < G(s") < 2 X lo4), in the temperature range of 10 to 25 "C. The anisotropy was more pronounced at the lower temperatures and was significantly pH dependent (Fig. 5). Analysis of the shear rate needed for the onset of anisotropy (13) indicated that the samples at pH 9.1 an+ 9.5 appeared to have persistence lengths of 800 to 1000 A at temperatures of 10 to 15°C (Table 11), consistent with the lengths determined using the contrast variation data and intensity data. These lengths dropped significantly when the temperature was raised to 25 "C. Fig. 5 indicates the level of anisotropy in scattering at different shear rates. These results show that although the persistence length of aggregates at pH 9.1 are similar at 10 and 15 "C, the sample at the lower temperature was significantly more aligned. Analysis of the sample at pH 7.5 showed that there was little alignment even at 15 "C. Presumably  The lengths were determined by measuring the shear rate needed to align the samples as described under "Experimental Procedures." The results assumed the aggregates were micelles with a diameter of 100-160 A, that the viscosity of the solution was 0.01 poise, and that T = 300 "K. and decreases at higher and lower pH. The intensity of scatter from samples at pH 9.1 under shear was fitted to the models of oriented cylinders (13).* To do so, the data were resolved into scatter perpendicular ( k l ) and parallel (kll) to the fluid flow. The data from samples at two rates of shear, 9,175 s" and 15,730 s-', were compared to the expected results for the model. The results (Fig. 6) indicate that the resolved data best fit the paired cylinder model and illustTates the closeness of .fit to the model of a tubular micelle 110 A in diameter, 1440 A long, and separated by 40 A, as derived from the static-scatt,ering studies.

DISCUSSION
The contrast variation results, the anisotropic-scattering patterns, and the Kratky plots of static and oriented samples are consistent with LPS at pH 9.1 cggregating as regular tubular micelles with lengths of 1200 A or more. The volume of the mice!les calculated from the contrast variation data is 1.37 X IO7 A3, while the model derived from the best fit of the Kratky plot of paired cylinders indicates a volume of 1.36 x 10' A". The smaller characteristic length at pH 9.1 Getermined from the Guinier analysis was approximately 40 A and indjcates that the micelles have diameters of about 80-100 ,A, consistent with the diameter suggested by the model (110 A). Such tubular micelles have been visualized in electron micrographs of LPS from smooth strains (8) as well as from rough strains in the Li salt form.3 As reported previously, the viscosity of the LPS solution at pH 9.1 was high and dropped at higher and lower pH values (7). Furthermore, the anisotropicscattering data indicated that the stiffness of the micelles increased as the temperature dropped from 25 to 1O"C, due perhaps to an ordering of the acyl chains within the micelles. For the sample at pH 9.1, the static scattering intensity at 25 "C and the resolved anisotropic scattering at 10 "C was unique and suggested that the micelles were paired. We propose that this pairing is the result of the tubular micelles closing back on themselves, forming hairpin-like micelles with both ends closed. The sides of such a loop would be the paired units. Such closed tubular micelles of LiLPS from an Ra strain have been seen in the electron microscope? The forces that stabilize the micellar form at pH 9.1 are thought to be the high head group charge repulsion and the steric hinderance of the hydrated head group region when the charge is high. Above p H 7.5, the second ionization of the two phosphates causes a dramatic increase in the negative charge of the lipid A head group. Charge repulsion and higher levels of bound water may induce an increase in head group spacing and, thus, in curvature of the aggregate. Since the LPS head group has a long and short axis (l), LPS may not readily pack into a spherical micelle; thus, the tubular micelle would be the stable form under these conditions. However, lipid at the end of a tubular micelle would pack as a hemisphere. If spherical packing is unstable at pH 9.1 due to the geometry of the molecules, ends of tubes would tend to coalesce, forming either longer tubes or circularly closed tubes.
At the lower pH values, 6.0 and 7.5, the scattering was consistent with floppy tubes of larger average diameters than at the higher pH. Such floppy tubes did not align in the shear cell, presumably due to the flexibility or lack of asymmetry in the aggregates. This type of irregular tube or ribbon has been visualized in electron micrographs of NaLPS from rough strains (6). As reported previously, the samples at neutral pH were cloudy compared to the clear samples at basic pH (7). This difference in light scattering also reflects differences in aggregate size.
When the pH of the sample was raised above 9.1, the angleaveraged scattering pattern was no longer consistent with paired tubes. The sample remained clear and still readily aligned in the shear cell. However, the viscosity of the sample at pH 9.5 was significantly lower than at pH 9.1. We propose that further increasing the charge on the LPS at pH values above 9.1 further increases the head group spacing within the lipid aggregates. This may allow for the formation of hemispherical micelles at the ends of the tubes. If the driving force to form closed loops and longer tubes is lessened, the micelles may then become smaller and unpaired. This is what occurred at pH 9.5. The anisotropic-scattering data indicated that the length of the micelles at pH 9.5 was shorter than at 9.1 ( Table  11). The decrease in the length of the micelles may account for the decrease in viscosity. We reiterate that the mere observation of anisotropic scattering, in our experimental geometry and at the shear rates used, rules out lamellar structures and globular structures with axial ratios less than about lO:l, regardless of any specific model features.
In conclusion, our results indicate that LPS from an Re strain can adopt several different micellar forms in the basic pH range. These conformational changes appear to be driven by charge interactions within and between the head groups of the molecules. Similar charge group interactions are likely to occur between LPS molecules from other rough as well as smooth strains (7). The formation of nonlamellar aggregates ' E. J. McGroarty, unpublished observations. Neutron Scattering Studies of LPS of LPS on the surface of the bacterial cell may also occur under specific conditions and may account for the release of LPS from bacterial cells.